Open Parisnewsy opened 8 months ago
Hi,
We are setting encoder freeze to be False
by default, which means the encoder is trainable.
We tried both freezing encoder and trainable encoder at earlier experiments and found that a trainable encoder reaches better loss and relatively better performances.
Hi OpenLRM Team I don't find encoder freeze in LRM paper, why do you choose to set freeze encoder is false?